Atlas uses a person’s own MRI to guide which implant contacts a decoder should listen to. Before applying that guidance, it checks whether it is appropriate for that person. If any check fails, Atlas records why and uses every contact instead. It works alongside the decoder you already use.
Evaluate Atlas on your data ↓ Run the checks yourself →This is a simplified view. See the preprint for the full applicability checks.
Atlas is designed to work alongside different neural decoders when a task-relevant brain area can be identified using the person's imaging. Potential applications include motor and speech decoding. This study tested speech versus music classification.
Run the checks yourself on any subject →Scores show balanced accuracy for speech versus music classification, averaged across calibration sets of 5 to 50 two-second windows. Chance performance is 50%.
These are exploratory results. This dataset is where Atlas showed the strongest results. In two other public datasets, visual MEG and visual ECoG, it showed no consistent advantage. The preprint reports all three.
These findings come from a small, exploratory study using a public dataset. The next step is to test Atlas on the remaining 8 subjects from the same dataset, with an analysis plan published before analyzing the data.
We also plan to evaluate a version of Atlas that adjusts how strongly imaging guidance is applied based on confidence, rather than simply applying or withholding it. This approach is patent pending.
The study results are documented in a public evidence package, and selected findings were also reproduced using separately written code.
Building neural interfaces? Start with a retrospective evaluation of data you already have. No new patient recruitment is needed. Atlas works alongside your existing decoder and can run in-house on your systems, so your data stays with you.
Or connect on LinkedIn.